application of positive matrix factorization in estimating

HAL Id: hal-00301671https://hal.archives-ouvertes.fr/hal-00301671

Submitted on 28 Jul 2005

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Application of positive matrix factorization inestimating aerosol secondary organic carbon in Hong

Kong and insights into the formation mechanismsZ. B. Yuan, J. Z. Yu, A. K. H. Lau, P. K. K. Louie, J. C. H. Fung

To cite this version:Z. B. Yuan, J. Z. Yu, A. K. H. Lau, P. K. K. Louie, J. C. H. Fung. Application of positive matrixfactorization in estimating aerosol secondary organic carbon in Hong Kong and insights into the for-mation mechanisms. Atmospheric Chemistry and Physics Discussions, European Geosciences Union,2005, 5 (4), pp.5299-5324. �hal-00301671�

https://hal.archives-ouvertes.fr/hal-00301671

https://hal.archives-ouvertes.fr

ACPD5, 5299–5324, 2005

Aerosol secondaryorganic carbon in

Hong Kong

Z. B. Yuan et al.

Title Page

Abstract Introduction

Conclusions References

Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version

Interactive Discussion

EGU

Atmos. Chem. Phys. Discuss., 5, 5299–5324, 2005www.atmos-chem-phys.org/acpd/5/5299/SRef-ID: 1680-7375/acpd/2005-5-5299European Geosciences Union

AtmosphericChemistry

and PhysicsDiscussions

Application of positive matrixfactorization in estimating aerosolsecondary organic carbon in Hong Kongand insights into the formationmechanismsZ. B. Yuan1, J. Z. Yu2, A. K. H. Lau1, P. K. K. Louie4, and J. C. H. Fung3

1Atmospheric, Marine and Coastal Environment Program, The Hong Kong University ofScience and Technology, Clear Water Bay, Hong Kong, China2Department of Chemistry, The Hong Kong University of Science and Technology, Clear WaterBay, Hong Kong, China3Department of Mathematics, The Hong Kong University of Science and Technology, ClearWater Bay, Hong Kong, China4Environmental Protection Department of HKSAR Government, 33/F, Revenue Tower, 5Gloucester Rd., Wanchai, Hong Kong, China

Received: 19 May 2005 – Accepted: 27 June 2005 – Published: 28 July 2005

Correspondence to: J. Z. Yu ([email protected])

© 2005 Author(s). This work is licensed under a Creative Commons License.

5299

http://www.atmos-chem-phys.org/acpd.htm

http://www.atmos-chem-phys.org/acpd/5/5299/acpd-5-5299_p.pdf

http://www.atmos-chem-phys.org/acpd/5/5299/comments.php

http://www.copernicus.org/EGU/EGU.html

ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Abstract

Secondary organic carbon (SOC) is often a significant portion of organic carbon (OC)in ambient particulate matter (PM). The levels and seasonal patterns of SOC in HongKong were examined using more than 2000 PM10 measurements made over a 4.5-yearperiod (1998–2002) in a network of ten air quality monitoring stations. The positive5

matrix factorization (PMF) model was used to analyze this large data set for sourceidentification and apportioning. SOC was subsequently estimated to be the sum ofOC present in the secondary sources, i.e., secondary sulfate, secondary nitrate, andsecondary organic aerosol. The annual average SOC as estimated by the PMF methodwas 4.25µg C/m3 while the summer average was 1.66µg C/m3 and the winter average10

was 7.05µg C/m3. In comparison, the method that uses EC as a tracer for primarycarbonaceous aerosol sources to derive SOC overestimated SOC by 70–212% for thesummer samples and by 4–43% for the winter samples. The overestimation by theEC tracer method resulted from the inability of obtaining a single OC/EC ratio thatrepresented a mixture of primary sources varying in time and space.15

We found that SOC and secondary sulfate had synchronous seasonal variation andwere correlated in individual seasons, suggesting common factors that control theirformation. Considering the well-established fact that both gas phase oxidation andin-cloud processing are important formation pathways for sulfate, the synchronicity ofSOC and sulfate suggests that in-cloud pathways are also important for SOC forma-20

tion. Additionally, the presence of SOC was found to be enhanced more than that ofsecondary sulfate in the winter. We postulate this to be a combined result of favorablepartitioning of semivolatile SOC species in the particle phase and more abundant SOCprecursors in the winter.

5300





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

1. Introduction

Carbonaceous aerosol consists of particulate elemental carbon (EC) and organic car-bon (OC). EC is formed through incomplete combustion of carbonaceous fuels, suchas diesel, gasoline, and organic wastes, while OC has two formation pathways, pri-mary and secondary. Primary OC is directly emitted from combustion sources while5

secondary organic carbon (SOC) is formed through atmospheric oxidation of reactiveorganic gases followed by gas-to-particle conversion processes. SOC is often a signif-icant portion of OC. Knowledge of the relative contribution of primary and secondaryOC is important in formulating effective control measures for ambient particulate matter(PM). Quantification of SOC is, however, difficult because of our limited understanding10

of the molecular composition of SOC and because of the presence of a large but un-known number of individual secondary organic products. Indoor and outdoor smogchamber experiments have identified numerous secondary organic compounds froma few major SOC precursors, including aromatic species (e.g., Forstner et al., 1997;Jang and Kamens, 2001), cyclic alkenes (Kalberer et al., 2000), and terpenes (e.g., Yu15

et al., 1999). To identify individual secondary organic compounds at a particular sitein ambient environment is, however, difficult, if not impossible. At present, no directmeasurement methods are available for determining SOC in atmospheric particulates.Consequently, indirect methods must be employed for the estimation of relative contri-butions of primary and secondary OC.20

Three indirect approaches to SOC estimation are found in the literature. In the firstapproach, EC is adopted as a “tracer” for calculating the abundance of primary OCbased on EC being exclusively primary in origin and on EC and OC having commonemission sources (e.g., Turpin and Huntzicker, 1995; Castro et al., 1999; Lim andTurpin, 2002). A ratio of OC/EC that is characteristic of primary emissions, called25

(OC/EC)p hereafter, is used to estimate SOC. Ambient OC/EC exceeding the (OC/EC)pratio is attributed to SOC. In the second approach, chemical transport models areused for calculating secondary organic aerosol yields by simulating formation, removal,

5301





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

and partitioning processes of SOC that are derived from smog chamber studies (e.g.,Bowman et al., 1997; Pandis et al., 1992; Strader et al., 1999). In the third approach,an organic tracer-based receptor model, Chemical Mass Balance (CMB), is used toapportion primary OC to a number of perceived sources, and the difference betweenthe measured total OC and the apportioned primary OC from the few known sources is5

regarded to be SOC (Schauer et al., 1996; Zheng et al., 2002; Na et al., 2004; Sawantet al., 2004).

All three approaches have caveats. In the first approach, the accuracy of SOC es-timates is highly dependent on the accuracy of the (OC/EC)p ratio in representing themixture of multiple local primary PM sources. The (OC/EC)p value varies among com-10

bustion sources. Consequently, it is temporally and spatially unique and influenced bymeteorology and diurnal and seasonal fluctuations in emissions. It is not a trivial matterto obtain a representative (OC/EC)p ratio for any given location (Strader et al., 1999;Seinfeld and Pankow, 2003). In practice, the (OC/EC)p ratio is often estimated fromambient measurement data by assuming either that the lowest observed OC/EC value15

represents the value for primary aerosols or that the average OC/EC value of the sam-ples collected on days of reduced photochemical activity reflects the (OC/EC)p value.The (OC/EC)p values obtained in such a manner, therefore, tend to be biased towardthe primary source(s) with relatively lower (OC/EC)p value and do not represent theaverage overall emission characteristics of the primary sources. Furthermore, in this20

method, OC in primary sources with low-EC or non-EC (e.g., cooking fume, biomassburning) is counted as SOC. The collective consequence is that the SOC levels areoverestimated to an unknown degree. Nevertheless, because of its simple nature andno need for measurement of other aerosol components, the EC tracer approach hashad widespread and continuous applications (e.g., Castro et al., 1999; Lin and Tai,25

2001; Viidanoja et al., 2002).The transport modeling approach is inherently limited by our knowledge of SOC for-

mation pathways and other atmospheric processes involving SOC. For instance, a fewrecent studies have indicated that in-cloud processing might be a significant pathway

5302





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

to form SOC (Warneck, 2003; Blando and Turpin, 2000). However, the reaction mech-anisms of in-cloud processes are largely unknown, let alone being represented quan-titatively in models. Large uncertainties also occur in the representation of emissioninventories of both gas- and particle-phase species and in the numerical representationof atmospheric dynamics. As a result, SOC estimation by transport models deviates5

from the actual values also to an unknown degree.The third approach (i.e., the CMB modeling approach) is also limited by the uncer-

tainties in the emission profiles. In addition, it is difficult to ensure that all the PMsources of statistical significance have been included in the model, either because ofhigh cost in establishing emission profiles for the numerous PM sources or because of10

failure in identifying all the important sources. The SOC thus obtained would also likelybe overestimated as a result of mistaking some unidentified sources for SOC.

In this paper, we describe a new approach for estimating SOC by using an advancedreceptor model, positive matrix factorization (PMF). PMF modeling of speciated PM10concentration data is carried out to identify source categories by collectively consider-15

ing component abundance in the derived source profiles and the corresponding tempo-ral variations in source contributions. SOC is subsequently taken to be the sum of OCin the sources of secondary nature. In comparison with the traditional receptor modelssuch as CMB, PMF does not require any a priori information on the source profilesas input, thus avoiding the uncertainties inherent in the source characterizations and20

limits imposed by a lack of known source profiles. Using extensive data from air qualitymonitoring network in Hong Kong, we compare SOC estimates by the PMF methodwith those by the EC tracer method to highlight the shortcomings of the latter method.Finally, the formation mechanism of SOC is discussed on the basis of its temporalvariation and its correlation with secondary sulfate.25

5303





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

2. Monitoring sites and sample analysis

The Hong Kong Environment Protection Department (HKEPD) has routinely conductedfilter-based 24-h PM10 sampling at ten air quality monitoring stations (AQMSs) on aschedule of once every six days since July 1998. The network of AQMSs that werechosen for filter-based PM10 sampling and subsequent detailed chemical speciation5

includes a roadside station (MK) and nine general stations (CW, KC, KT, SSP, ST, TC,TP, TW, and YL). The spatial distribution of the ten AQMSs is illustrated in Fig. 1. Adetailed description of the site characteristics is presented by Yu et al. (2004). Sampleswere collected onto 8×10 in quartz fiber filters using high volume PM10 samplers oper-ating at a flow rate of 1.13 m3/min. Twenty-six elements and water-soluble ions (Al, As,10

Ba, Be, Br−, Ca, Cd, Cl−, Cr, Cu, EC, Fe, Hg, K+, Mg, Mn, Na+, NH+4 , Ni, NO−

3 , OC, Pb,

Se, SO2−4 , V and Zn) were determined for the samples. Among them, the ionic species

were measured using ion chromatography. As, Hg and Se were measured using aflow injection analysis – atomic absorption technique and all the other elements weremeasured using inductively coupled plasma atomic emission spectroscopy. EC and15

OC were analyzed by a thermal/optical transmittance method following the protocol ofNIOSH method 5040 (Sin et al., 2002). The PM10 analyses were subjected to strict in-house quality assurance/quality control (QA/QC) procedures described in the report bythe HKEPD (2002). Typically, the control limits for the PM10 measurements were ±10%for both accuracy and precision. A speciated data set of 2199 samples spanning from20

July 1998 to December 2002 was used in this study to examine the abundance andseasonal patterns of SOC.

5304





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

3. SOC estimation

3.1. EC tracer method

For comparison purposes, we also estimated SOC by the EC tracer method. In thismethod, SOC is calculated as follows:

SOC=OC−EC(OC/EC)p. (1)5

The relative contributions of various aerosol sources were expected to be season-dependent. Consequently, the representative (OC/EC)p is also season-dependent. Forthis reason, season-specific (OC/EC)p values were approximated using samples thathad the lowest 5% measured (OC/EC) values in a given season. In accord with thesubtropical climate in Hong Kong, the seasons were defined as follows: spring from10

16 March to 15 May; summer from 16 May to 15 September; fall from 16 Septemberto 15 November and winter from 16 November to 15 March of the next year (Chin,1986). Spring and fall are two short and transient seasons, lasting approximately twomonths. A least-square regression of the OC and EC data in the lowest 5% of OC/ECratio yielded a slope of 0.73 (r2=0.99, n=17), 0.41 (r2=0.90, n=39), 0.70 (r2=0.92,15

n=20), and 0.88 (r2=0.92, n=34), for spring, summer, fall, and winter measurements,respectively (Fig. 2).

3.2. The PMF source apportionment method

Receptor models estimate the amount of pollutants contributed by different sourcesthrough statistical interpretation of ambient observation data and reasonably imposed20

constraints. A detailed explanation of the working principle of PMF can be found in thePMF user’s guide (Paatero, 2000). Our PMF analysis of the 4.5-year dataset identifiednine significant aerosol sources, comprising vehicle exhaust, residual oil, fresh sea salt,aged sea salt, crustal soil, regional combustion sources (e.g., biomass burning), sec-ondary sulfate, secondary nitrate, and secondary organic aerosol (SOA). The source25

5305





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

categories were identified by the explained variations in species in individual appor-tioned sources as well as the temporal and spatial variation of source contributions.The source identification procedure for this data set was described in detail by Yuan etal. (20051, and available at http://landsea.ust.hk/∼chjianyu/Yu Papers/PMF paper.pdf).Table 1 lists the resolved source categories and their profiles. OC in the three sec-5

ondary aerosol source categories, i.e., secondary sulfate, secondary nitrate, and sec-ondary organic aerosol, was summed up to represent SOC.

4. Results and discussion

4.1. Comparison of SOC derived from the EC tracer method and the PMF method

The two plots in Fig. 3 compare the average SOC levels estimated by the two methods10

at individual AQMSs in the summer and in the winter, respectively. Due to the short andtransient nature of spring and fall in Hong Kong, our discussion of SOC abundance andseasonality is focused on the summer and winter. Figure 3 clearly shows that the ECtracer method consistently gave higher SOC values, in line with the previous discussionthat the EC tracer method tends to overestimate the SOC level because of the inherent15

deviations from the actual (OC/EC)p.The shortcomings of the EC tracer method can be seen in the spatial pattern of

the deviation between the two sets of SOC estimates. The smallest difference of thetwo sets of SOC estimates occurred at TC (1.14µg/m3 in summer and 0.26µg/m3

in winter), a developing residential area with a smaller number of aerosol emission20

sources in comparison with other AQMSs. The two largest differences occurred at MK(4.72µg/m3 in summer and 2.95µg/m3 in winter), an urban roadside station, and KC(3.50µg/m3 in summer and 2.83µg/m3 in winter), a station in an industrialized area

1Yuan, Z. B., Lau, A. K. H., Zhang, H. Y., Yu, J. Z., Louie, P. K. K., and Fung, J. C. H.: Iden-tification and spatial-temporal variations of dominant PM10 sources over Hong Kong, Atmos.Environ., in review, 2005.

5306





http://landsea.ust.hk/~chjianyu/Yu_Papers/PMF_paper.pdf

ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

close to the city’s container terminal. In both locations, the primary sources of car-bonaceous aerosols were intensive and diversified, and their intensities were highlytime dependent. Such a spatial pattern in the difference between the two sets of SOCestimates could be attributed to the biased (OC/EC)p derived from EC and OC mea-surements of the lowest 5% OC/EC values. This bias was more prominent at locations5

with a larger number of significant sources that had diverse (OC/EC)p ratios. In a morestrict approach, station-specific (OC/EC)p should be derived considering the distinctlocal characteristics of the PM sources among individual AQMSs, instead of lumpingtogether measurements from different stations and then implementing a uniform valueof (OC/EC)p. However, the number of measurements at individual stations is not large10

enough to support this approach.Next we examine the seasonal pattern in the differences of the two sets of SOC

estimates. In the summer, the difference among the ten AQMSs ranged from 1.1to 4.3µg C/m3, whereas in the winter, the difference was smaller, ranging from 0.3to 2.9µg C/m3. In our study, the derived summer (OC/EC)p was 0.41, significantly15

smaller than reported (OC/EC)p values in the literature, which range from 0.74 in theurban area of Nagoya, Japan (Kadowaki, 1990) to 2.4 in Pittsburgh, PA, U.S. (Cabadaet al., 2002). This derived (OC/EC)p value of 0.41 is at the lower end of the rangeof (OC/EC)p measurements for sources aerosols emitted from combustion of heavy-duty diesel vehicles. Gillies and Gertler (2000) evaluated four databases of source20

profiles in the U.S. and concluded that diesel vehicle exhausts had average OC/ECratios ranging from 0.28 to 0.92. Diesel-fueled vehicles account for 25% of registeredvehicles in Hong Kong, and the most-traveled vehicles such as public buses accountfor the majority of diesel-fueled vehicles (HKTD, 2002). In addition, the ships at thecontainer ports mainly burn residual oil. The summer (OC/EC)p estimate is most likely25

biased towards diesel and residual oil combustion aerosols, which typically have lower(OC/EC)p than other combustion aerosols. This would at least be partially responsiblefor the significant overestimation of summer SOC by the EC tracer method.

A second reason for the overestimate of summer SOC by the EC tracer method

5307





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

is counting non-EC associated OC as SOC. This is revealed in Fig. 4, which plots theSOC estimates by the EC method versus those by the PMF method in the summer andin the winter. The slopes of both the summer and the winter regression lines are largerthan 1, confirming the overestimate of SOC by the EC tracer method. The summer plothas a large intercept (1.64), whereas the winter plot has a near-zero intercept (−0.36).5

The intercept values reflect that the portion of primary OC (e.g., biogenic OC) with noassociated EC was more prominent in the summer than in the winter. A larger presenceof biogenic OC in the summer PM2.5 in Hong Kong was reported by Zheng et al. (2000)using Carbon Preference Index (CPI) for n-alkanes, n-fatty acids, and n-alkanols.

An additional observation is that the summer plot is more scattered (r2=0.48,10

n=785), in comparison with the winter plot (r2=0.77, n=687) (Fig. 4). The more scat-tered summer data could be explained if the significant primary sources in the summerwere more diverse in (OC/EC)p. The EC tracer method was hardly able to capture thereal and varied (OC/EC)p by using the lowest 5% standard or any other single-valuestandards.15

In summary, the diversity of (OC/EC)p among significant primary sources and thepresence of primary OC sources without EC pose significant limitations in the use ofthe EC tracer method for estimating SOC. In comparison, PMF identifies the profilefor each source or a group of sources with similar characteristics and calculates thesource impact for each sample. That is, the PMF method does not make the unrealistic20

assumption of a uniform (OC/EC)p on different sampling dates in the same season. Wetherefore believe that the PMF method yields estimates closer to the true SOC values.For this reason, the PMF-derived SOC results are used in the ensuing discussion. Thelimitation to the PMF approach is that a large speciated PM data set is a prerequisiteto enable source apportioning. Obtaining such a level of comprehensive measurement25

data requires a PM monitoring network running routinely for an extended period of time(e.g., a year) and analytical capabilities for various aerosol constituents.

5308





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

4.2. Abundance and seasonality of SOC

The average SOC level over the 4.5-year period for Hong Kong was 4.25µg C/m3,contributing 46% of the aerosol OC loadings in the air. On a material mass basis, thesecondary organic aerosol contributed 13% of PM10 mass, if we use a multiplier of 1.6to convert SOC to SOA mass (Turpin and Lim, 2001). This implies the importance of5

identifying SOC precursors for effective reduction of aerosol loadings.The SOC levels showed a clear seasonal pattern of being higher in the winter and

lower in the summer (Fig. 3). The winter average SOC was 7.05µg C/m3, nearly fourtimes the summer average of 1.66µg C/m3. In comparison, primary OC had a winteraverage of 6.17µg C/m3, 1.6 times the summer average of 3.72µg C/m3. Apparently,10

SOC increased more than the primary OC in the winter. As a result, the relative contri-bution of SOC to OC was higher by 70% in the winter than in the summer.

The seasonal pattern for SOC in Hong Kong was in contrast to that reported inmany other locations in the world, where typically higher SOC was observed duringthe summer by virtue of more conducive conditions (e.g., clear skies and low winds)15

and enhanced photochemical activities. The locations where higher SOC was reportedin the summer than in the winter include Birmingham in England (Castro et al., 1999),Pittsburgh, PA (Cabada et al., 2002), Southern California (Strader et al., 1999; Turpinand Huntzicker, 1995; Chu and Macias, 1981), and St. Louis, MO (Chu and Macias,1981) in the U.S. The unusual seasonal pattern of SOC in Hong Kong can be explained20

by the meteorological conditions unique to Hong Kong. During the summer in HongKong, frequent rainfall effectively removes aged aerosols. Located in the subtropicalregion, Hong Kong has abundant sunlight and mild ambient temperatures (15–21◦C)even during the winter. Similar to Hong Kong, two southern European towns, Oportoand Coimbra in Portugal, where sunny winter days are frequent, also had higher SOC25

concentrations in the winter than in the summer (Castro et al., 1999).

5309





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

4.3. Relationship between SOC and secondary sulfate

Our PMF analysis has also identified secondary sulfate as a source category. Figure 5overlays the temporal variation of secondary sulfate with that of SOC at six AQMSswith the least uninterrupted monitoring data. At all the monitoring stations, includ-ing the roadside station, the seasonal variations of secondary sulfate and SOC were5

synchronous. Both exhibited higher concentrations in the winter than in the summer.Figure 6 plots secondary sulfate versus SOC for the summer and the winter samples.The two species were well correlated in individual seasons, with r2 values of 0.76 and0.46 for the summer and the winter samples, respectively. The close tracking of the twosuggests that a same set of conditions predominantly affect the abundance of the two10

species. It is well established that both gas-phase and in-cloud oxidation contribute tothe formation of sulfate (Seinfeld and Pandis, 1998). It is also increasingly recognizedthat SOC can be formed in in-cloud oxidation processes in addition to through gasphase oxidation processes (Warneck, 2003; Blando and Turpin, 2000; Yu et al., 2005).However, the reaction pathways, mechanisms, and the corresponding thermodynamic15

coefficients for SOC formation through in-cloud processing remain largely unknown.An additional observation is that the summer samples are more concentrated in the

upper part of the figure, while most of the winter samples are located in the lower part.This is reflected in the slope values (2.19 for the summer samples versus 0.72 for thewinter samples) of the sulfate-versus-SOC regression lines (Fig. 6). In other words,20

the presence of SOC was enhanced more than that of secondary sulfate in the winter.The winter average secondary sulfate concentration was 7.72µg/m3, about 2.5 timesthe summer average concentration of 3.20µg/m3. In comparison, the winter SOCenhancement over the summer SOC was more than 4 fold (7.05 versus 1.66µg C/m3).

We now suggest the underlying reasons for the seasonal differences in the relative25

abundance of the two secondary species. As a first-order approximation, the relative

5310





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

production ratio can be expressed by the following equation:

PSSPSOC

=kSS [Ox] [SO2]

αkSOC [Ox] [X], (2)

where kSS and kSOC are the “conceptual” rate constants for reactions leading to theformation of secondary sulfate and SOC, respectively; [SO2] and [X] are the concen-trations of SO2 and SOC precursors; [Ox] is the concentration of oxidants responsible5

for the oxidation of SO2 and X. Regardless of the nature of the formation pathways, ei-ther in the gas phase or in cloud, the SOC species undergoes gas-particle partitioningat the end. α is included in the equation to represent the mass fraction of the condens-able secondary organic species that partition into the aerosol phase. Conceivably, αis an inverse function of temperature. This has been demonstrated in a number of10

studies using smog chamber simulation or absorptive-partitioning models. For exam-ple, Takekawa et al. (2003) observed in their smog chamber experiments that the SOAyield at 283 K was approximately twice that at 303 K. In a modeling study, Sheehanand Bowman (2001) estimated that a 10◦C decrease in temperature would lead to anincrease of SOA yields by 20–150%, depending on the vaporization enthalpy. Hong15

Kong has a moderate and humid climate with a typical average daily temperature of∼30◦C in the summer and 15◦C in the winter. A significantly higher α value is an-ticipated in the winter than in the summer. On the other hand, sulfuric acid (H2SO4)has an extremely low vapor pressure and, consequently, sulfate resides in the aerosolphase once it is formed at the ambient temperature encountered either in the summer20

or in the winter. That is, the fraction of sulfate residing in the aerosol phase equals toone and is little influenced by the ambient temperature. The temperature affects therate constants, kSS and kSOC, in the same direction; however, it is difficult to estimatethe seasonal change in their relative magnitudes.

If we use the concentration of total non-methane hydrocarbon (TNMHC) as a rough25

indicator for the abundance of SOC precursors, the available hydrocarbon measure-ments in one annual cycle (August 2002–July 2003) indicate that the SOC precursorswere more abundant in the winter months (Fig. 7). In comparison, SO2 concentrations

5311





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

peaked in August, presumably linked to higher demands for energy for air conditioners(Fig. 7). In summary, the observation that the presence of SOC was enhanced morethan that of sulfate in the winter is most likely a combined result of the temperature-induced favorable partitioning of SOC into the aerosol phase and more abundant SOCprecursors.5

5. Conclusions

A SOC estimation method using PMF for source apportioning was introduced in thispaper. In PMF, OC apportioned in three secondary sources was summed to be SOC.The SOC estimates were compared with those obtained using the EC tracer methodfor a large dataset collected at ten AQMSs in Hong Kong from 1998 to 2002. The SOC10

estimates by the EC tracer method were consistently higher than those determinedby the PMF method. The difference in SOC estimates by the two methods rangedfrom 1.1–4.3µg C/m3 (70–212%) in the summer and 0.3–2.9µg C/m3 (4–43%) in thewinter. The comparison between the two sets of SOC estimates exposes the inherentshortcomings in the EC tracer method. The average SOC level estimated using PMF15

was 4.25µg C/m3, contributing nearly half of the carbonaceous aerosols in Hong Kong.The percentage contribution of SOC to OC was 46% on an annual average basis and38% and 52% in the summer and in the winter, respectively. On a material massbasis, the secondary organic aerosol contributed 13% of PM10 mass. Seasonally, thispercentage contribution was 7% in the summer months and 16% in the winter months.20

This implies the importance of identifying SOC precursors for effective reduction ofaerosol loadings.

SOC also showed a clear seasonal variation, with higher loading in the winter thanin the summer. The winter average (7.05µg C/m3) was approximately four times thesummer average (1.66µg C/m3). This seasonal pattern was similar to the variation of25

secondary sulfate concentrations. The two species correlated well in individual sea-sons, indicating common factors that control their formation. It has been established

5312





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

that sulfate has two major formation pathways, gas-phase oxidation by OH radicals andin-cloud oxidation processes. The close tracking of sulfate and SOC appears to sug-gest that the in-cloud pathway is also important for SOC formation. The slope of thesulfate-versus-SOC regression line, an indicator of the relative abundance of the two,for the summer samples was three times that for the winter samples. This is most likely5

a combined result of temperature-induced less favorable partitioning of semivolatilesecondary organics in the particle phase and the lower abundance of SOC precursorsin the summer.

Acknowledgements. The work was partially supported by the Research Grants Council ofHong Kong, China (602103). The authors thank the HKEPD for provision of the data sets10

and permission for publication.

References

Blando, J. D. and Turpin, B. J.: Secondary organic aerosol formation in cloud and fog droplets:a literature evaluation of plausibility, Atmos. Environ., 34, 1623–1632, 2000.

Bowman, F. M., Odum, J. R., and Seinfeld, J. H.: Mathematical model for gas-particle partition-15

ing of secondary organic aerosols, Atmos. Environ., 31, 3921–3931, 1997.Cabada, J. C., Pandis, S. N., and Robinson, A. L.: Sources of atmospheric carbonaceous

particulate matter in Pittsburgh, Pennsylvania, J. Air & Waste Manage. Assoc., 52, 732–741,2002.

Castro, L. M., Pio, C. A., Harrison, R. M., and Smith, D. J. T.: Carbonaceous aerosol in urban20

and rural European atmospheres: estimation of secondary organic carbon concentrations,Atmos. Environ., 33, 2771–2781, 1999.

Colman, J. J., Swanson, A. L., Meinardi, S., Sive, B. C., Blake, D. R., and Rowland, F. R.: De-scription of the analysis of a wide range of volatile organic compounds in whole air samplescollected during PEM-Tropics A and B; Anal. Chem., 73, 3723–3731, 2001.25

Chin, P. C.: Climate and weather, in: A Geography of Hong Kong, edited by: Chiu, T. N. andSo, C. L., Oxford University Press, New York, pp. 69–85, 1986.

Chu, L.-C. and Macias, E. S.: Carbonaceous urban aerosol – primary or secondary?, in: Atmo-

5313





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

spheric aerosol: source / air quality relationship, edited by: Macias, E. S. and Hopke, P. K.,ACS Sympoisum Series 167, American Chemical Society, Washington DC, 1981.

Crahan, K. K., Hegg, D., Covert, D. S., and Jonsson, H.: An exploration of aqueous oxalic acidproduction in the coastal marine atmosphere, Atmos. Environ., 38, 3757–3764, 2004.

Gillies, J. A. and Gertler, A. W.: Comparison and evaluation of chemically speciated mobile5

source PM2.5 particulate matter profiles. J. Air & Waste Manage. Assoc., 50, 1459–1480,2000.

Forstner, H. J. L., Flagan, R. C., and Seinfeld, J. H.: Secondary organic aerosol from thephotooxidation of aromatic hydrocarbons: molecular composition, Environ. Sci. Technol., 31,1345–1358, 1997.10

HKEPD (Hong Kong Environment Protection Department): Air services group, Air qual-ity in Hong Kong 2002, (http://www.epd.gov.hk/epd/english/environmentinhk/air/air quality/aq annualrpt.html).

HKTD (Hong Kong Transport Department): Monthly traffic and transport digest, available at:http://www.info.gov.hk/td/eng/publication/mtnt.htm, 2002.15

Jang, M. S. and Kamens, R. M.: Characterization of secondary aerosol from the photooxidationof toluene in the presence of NOx and 1-propene, Environ. Sci. Technol., 35, 3626–3639,2001.

Kadowaki, S.: Characterization of carbonaceous aerosols in the Nagoya urban area. 1. Ele-mental and organic carbon concentrations and the origin of organic aerosols, Environ. Sci.20

Technol., 24, 741–744, 1990.Kalberer, M., Yu, J., Cocker, D. R., Flagan, R. C., and Seinfeld, J. H.: Aerosol formation in the

cyclohexene-ozone system, Environ. Sci. Technol., 34, 4894–4901, 2000.Kawamura, K. and Ikushima, K.: Seasonal changes in the distribution of dicarboxylic acids in

the urban atmosphere, Environ. Sci. Technol., 27, 2227–2235, 1993.25

Lim, H. J. and Turpin, B. J.: Origins of primary and secondary organic aerosol in Atlanta:Results of time-resolved measurements during the Atlanta Supersite experiment, Environ.Sci. Technol., 36, 4489–4496, 2002.

Lin, J. J. and Tai, H. S.: Concentrations and distributions of carbonaceous species in ambientparticles in Kaohsiung City, Taiwan, Atmos. Environ., 35, 2627–2636, 2001.30

Na, K. S., Sawant, A. A., Song, C., and Cocker, D. R.: Primary and secondary carbonaceousspecies in the atmosphere of Western Riverside County, California, Atmos. Environ., 38,1345–1355, 2004.

5314





http://www.epd.gov.hk/epd/english/environmentinhk/air/air_quality/aq_annualrpt.html



http://www.info.gov.hk/td/eng/publication/mtnt.htm

ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Paatero, P.: User’s guide for positive matrix factorization programs PMF2 and PMF3, Part 1:tutorial, 2000.

Pandis, S. N., Harley, R. A., Cass, G. R., and Seinfeld, J. H.: Secondary organic aerosolformation and transport, Atmos. Environ., 26A, 2269–2282, 1992.

Sawant, A. A., Na, K. S., Zhu, X. N., and Cocker, D. R.: Chemical characterization of outdoor5

PM2.5 and gas-phase compounds in Mira Loma, California, Atmos. Environ., 38, 5517–5528,2004.

Schauer, J. J., Rogge, W. F., Hildemann, L. M., Mazurek, M. A., Cass, G. R., and Simoneit,B. R. T.: Source apportionment of airborne particulate matter using organic compounds astracers, Atmos. Environ., 30, 3837–3855, 1996.10

Seinfeld, J. H. and Pandis, S. N.: Atmospheric chemistry and physics: from air pollution toclimate change, John Wiley and Sons, New York, 1998.

Seinfeld, J. H. and Pankow, J. F.: Organic atmospheric particulate material, Annu. Rev. Phys.Chem., 54, 121–140, 2003.

Sheehan, P. E. and Bowman, F. M.: Estimated effects of temperature on secondary organic15

aerosol concentrations, Environ. Sci. Technol., 35, 2129–2135, 2001.Sin, D. W. M., Fung, W. H., and Lam, C. H.: Measurement of carbonaceous aerosols: validation

and comparison of a solvent extraction – gas chromatographic method and a thermal opticaltransmittance method, Analyst, 127, 614–622, 2002.

Strader, R., Lurmann, F., and Pandis, S. N.: Evaluation of secondary organic aerosol formation20

in winter, Atmos. Environ., 33, 4849–4863, 1999.Takekawa, H., Minoura, H., and Yamazaki, S.: Temperature dependence of secondary or-

ganic aerosol formation by photo-oxidation of hydrocarbons, Atmos. Environ., 37, 3413–3424, 2003.

Turpin, B. J. and Huntzicker, J. J.: Identification of secondary organic aerosol episodes and25

quantification of primary and secondary organic aerosol concentrations during SCAQS, At-mos. Environ., 29, 3527–3544, 1995.

Turpin, B. J. and Lim, H. J.: Species contributions to PM2.5 mass concentrations: revisiting com-mon assumptions for estimating organic mass, Aerosol Sci. Technol., 35, 602–610, 2001.

Viidanoja, J., Sillanpaa, M., Laakia, J., Kerminen, V. M., Hillamo, R., Aarnio, P., and Koskentalo,30

T.: Organic and black carbon in PM2.5 and PM10: 1 year of data from an urban site in Helsinki,Finland, Atmos. Environ., 36, 3183–3193, 2002.

Warneck, P.: In-cloud chemistry opens pathway to the formation of oxalic acid in the marine

5315





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

atmosphere, Atmos. Environ. 37, 2423–2427, 2003.Yu, J. Z., Cocker, D. R., Griffin, R. J., Flagan, R. C., and Seinfeld, J. H.: Gas-phase ozone

oxidation of monoterpenes: Gaseous and particulate products, J. Atmos. Chem., 34, 207–258, 1999.

Yu, J. Z., Huang, X.-F., Xu, J. H., and Hu, M.: When aerosol sulfate goes up, so does oxalate:5

implication for the formation mechanisms of oxalate, Environ. Sci. Technol., 39, 128–133,2005.

Yu, J. Z., Tung, J. W. T., Wu, A. W. M., Lau, A. K. H., Louie, P. K. K., and Fung, J. C. H.:Abundance and seasonal characteristics of elemental and organic carbon in Hong KongPM10, Atmos. Environ., 38, 1511–1521, 2004.10

Zheng, M., Fang, M., Wang, F., and To, K. L.: Characterization of the solvent extractable organiccompounds in PM2.5 aerosols in Hong Kong, Atmos. Environ., 34, 2691–2702, 2000.

Zheng, M., Cass, G. R., Schauer, J. J., and Edgerton, E. S.: Source apportionment of PM2.5in the southeastern United States using solvent-extractable organic compounds as tracers,Environ. Sci. Technol., 36, 2361–2371, 2002.15

5316





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Table 1. Profiles of nine aerosol sources resolved by PMF analysis, values listed here aremass percentages in each source profile.

species Vehicle Residual Fresh Aged Crustal Secondary Secondary Regional SecondaryExhaust Oil Sea Salt Sea Salt Soil Sulfate Nitrate Combustion Organics

Al 0.00 0.21 0.01 0.00 11.74 0.16 0.00 0.01 0.51As 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.14Ca 0.87 6.05 0.97 0.33 30.93 0.00 0.19 1.83 1.18Cd 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.03Cl− 0.03 0.12 53.43 0.03 0.08 0.00 1.55 0.28 0.04Fe 1.22 0.10 0.00 0.00 15.77 0.39 0.68 0.44 1.67K+ 0.28 2.73 0.39 2.09 0.02 0.00 0.00 14.22 0.03Mg 0.00 0.01 3.03 3.02 5.43 0.00 0.00 0.00 0.00Mn 0.01 0.00 0.01 0.01 0.44 0.01 0.02 0.10 0.08Na+ 0.02 6.04 25.75 26.90 0.02 0.01 0.01 0.31 0.03NH4+ 0.00 0.17 0.01 0.01 0.01 15.74 8.68 0.04 0.07Ni 0.00 1.34 0.00 0.00 0.02 0.00 0.00 0.00 0.00NO−

3 0.04 0.07 0.54 3.74 2.59 0.00 11.80 0.01 0.01Pb 0.00 0.00 0.00 0.00 0.03 0.04 0.08 0.84 0.92SO=

4 2.36 12.86 0.06 59.43 27.11 61.47 0.05 15.92 10.84V 0.00 2.80 0.00 0.01 0.00 0.00 0.00 0.02 0.00EC 40.35 42.11 10.36 4.34 5.34 7.42 2.31 0.23 0.19OMa 54.81 25.39 5.45 0.09 0.46 14.74 74.63 65.73 84.28

a OM stands for Organic Matter defined as 1.6 times Organic Carbon (OC) concentrations inthis study (Turpin and Lim, 2001).

5317





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Guangdong,Mainland China

NewTerritories

Kowloon

Hong KongIslandLantau

NNHong Kong

MKRoadside Station

YL

TW

CW

General AQ Station

TP

ST

KCSSP

KT1 km

TC

Fig. 1. Location map of the air quality monitoring stations in Hong Kong that were selected forfilter-based PM10 sampling and subsequent speciation analysis. (The abbreviations are: CW(Central/Western), KC (Kwai Chung), KT (Kwun Tong), MK (Mong Kok), SSP (Sham Shui Po),ST (Shatin), TC (Tung Chung), TP (Tai Po), TW (Tsuen Wan), and YL (Yuen Long).)

5318





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Fig. 2. Scatter plots of OC versus EC for samples of the lowest 5% (OC/EC) ratios.

5319





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Summer

0

4

8

12

16

20

CW KC KT MK SSP ST TC TP TW YL

Con

cent

ratio

n(µ

gC/m

3) EC Tracer

PMF

Winter

0

4

8

12

16

20

CW KC KT MK SSP ST TC TP TW YL

Con

cent

ratio

n(µ

gC/m

3)

EC Tracer

PMF

Fig. 3. Comparison of SOC estimates by the EC tracer method and by the PMF method.

5320





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

y = 1.30x + 1.64R2 = 0.48

0

5

10

15

20

25

30

0 3 6 9 12 15

SOC estimation by PMF (µgC/m3)

SOC

est

imat

ion

by E

C tr

acer

(µgC

/m3 )

y = x

Summer

y = 1.28x - 0.36R2 = 0.77

0

10

20

30

40

50

0 10 20 30 40SOC estimation by PMF (µgC/m3)

SOC

est

imat

ion

by E

C tr

acer

(µgC

/m3 )

y = x

Winter

Fig. 4. Scatter plots between SOC estimates by the EC tracer method and the PMF methodfor the summer samples (Top) and the winter samples (Bottom).

5321





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Con

cent

ratio

n(µ

g/m

3)

SOC

secondary sulfateCW

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Co

nce

ntr

atio

n(µ

g/m

3)

SOC

secondary sulfateKT

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Co

nce

ntr

atio

n(µ

g/m

3)

SOC

secondary sulfateMK

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2C

on

cen

tra

tion

(µg

/m3)

SOC

secondary sulfateSSP

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Co

nce

ntr

atio

n(µ

g/m

3)

SOC

secondary sulfateTW

0

10

20

30

40

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Co

nce

ntr

atio

n(µ

g/m

3)

SOC

secondary sulfateYL

Fig. 5. Time series of SOC and secondary sulfate at six air quality monitoring stations.

5322





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

y = 2.19x - 0.42R2 = 0.76

y = 0.72x + 2.63R2 = 0.46

0

10

20

30

40

0 10 20 30 40SOC(µgC/m3)

Sec

Sul

fate

(µg/

m3 )

WinterSummer

Fig. 6. Scatter plots of SOC versus secondary sulfate.

5323





ACPD5, 5299–5324, 2005


Hong Kong

Z. B. Yuan et al.

Title Page



Tables Figures

J I

J I

Back Close

Full Screen / Esc

Print Version


EGU

Fig. 7. Monthly average concentrations of SO2 and the total nonmethane hydrocar-bon (TNMHC) at an urban air quality monitoring station (Central/Western) in Hong Kongfor the period of August 2002–July 2003. (TNMHC is the sum of 39 species compris-ing ethane, ethane, ethyne, propane, propene, i-butane, n-butane, 1-butene, i-butene,trans-2-butene, cis-2-butene, i-pentane, n-pentane, 1,3-butadiene, 1-pentene, isoprene, 2,3-dimethylbutane, 2-methylpentane, 3-methylpentane, n-hexane, 2,2,4-trimethylpentane, 2-methylhexane, 3-methylhexane, n-heptane, n-octane, benzene, toluene, ethylbenzene, m-xylene, p-xylene, o-xylene, isopropylbenzene, propylbenzene, 3-ethlytoluene, 4-ethyltoluene,1,3,5-trimethylbenzene, 2-ethyltoluene, 1,2,4-trimethylbenzene, and 1,2,3-trimethylbenzene.The chemical analysis was done by Professor Donald Blake’s group at the University ofCalifornia-Irvine. We refer readers to Colman et al. (2001) for the methodology and the QA/QCdetails.)

5324





application of positive matrix factorization in estimating

Documents